visualization2_with_ggplot2
Emily Potts
2022-10-04
weather_df =
rnoaa::meteo_pull_monitors(
c("USW00094728", "USC00519397", "USS0023B17S"),
var = c("PRCP", "TMIN", "TMAX"),
date_min = "2017-01-01",
date_max = "2017-12-31") %>%
mutate(
name = recode(
id,
USW00094728 = "CentralPark_NY",
USC00519397 = "Waikiki_HA",
USS0023B17S = "Waterhole_WA"),
tmin = tmin / 10,
tmax = tmax / 10) %>%
select(name, id, everything())## Registered S3 method overwritten by 'hoardr':
## method from
## print.cache_info httr## using cached file: ~/Library/Caches/R/noaa_ghcnd/USW00094728.dly## date created (size, mb): 2022-09-29 10:31:53 (8.401)## file min/max dates: 1869-01-01 / 2022-09-30## using cached file: ~/Library/Caches/R/noaa_ghcnd/USC00519397.dly## date created (size, mb): 2022-09-29 10:32:02 (1.699)## file min/max dates: 1965-01-01 / 2020-03-31## using cached file: ~/Library/Caches/R/noaa_ghcnd/USS0023B17S.dly## date created (size, mb): 2022-09-29 10:32:05 (0.95)## file min/max dates: 1999-09-01 / 2022-09-30weather_df## # A tibble: 1,095 × 6
## name id date prcp tmax tmin
## <chr> <chr> <date> <dbl> <dbl> <dbl>
## 1 CentralPark_NY USW00094728 2017-01-01 0 8.9 4.4
## 2 CentralPark_NY USW00094728 2017-01-02 53 5 2.8
## 3 CentralPark_NY USW00094728 2017-01-03 147 6.1 3.9
## 4 CentralPark_NY USW00094728 2017-01-04 0 11.1 1.1
## 5 CentralPark_NY USW00094728 2017-01-05 0 1.1 -2.7
## 6 CentralPark_NY USW00094728 2017-01-06 13 0.6 -3.8
## 7 CentralPark_NY USW00094728 2017-01-07 81 -3.2 -6.6
## 8 CentralPark_NY USW00094728 2017-01-08 0 -3.8 -8.8
## 9 CentralPark_NY USW00094728 2017-01-09 0 -4.9 -9.9
## 10 CentralPark_NY USW00094728 2017-01-10 0 7.8 -6
## # … with 1,085 more rowsscatterplot, but better this time:
weather_df %>%
ggplot(aes(x = tmin, y = tmax)) +
geom_point(aes(color = name), alpha = .5)## Warning: Removed 15 rows containing missing values (geom_point).
labels:
weather_df %>%
ggplot(aes(x = tmin, y = tmax)) +
geom_point(aes(color = name), alpha = .5) +
labs(
title = "Temperature plot",
x = "Minimum daily temperature (C)",
y = "Maxiumum daily temperature (C)",
caption = "Data from the rnoaa package"
)## Warning: Removed 15 rows containing missing values (geom_point).
scales:
weather_df %>%
ggplot(aes(x = tmin, y = tmax)) +
geom_point(aes(color = name), alpha = .5) +
labs(
title = "Temperature plot",
x = "Minimum daily temperature (C)",
y = "Maxiumum daily temperature (C)",
caption = "Data from the rnoaa package") +
scale_x_continuous(
breaks = c(-15, 0, 15),
labels = c("-15º C", "0", "15"))## Warning: Removed 15 rows containing missing values (geom_point).
Aesthetic mappings determine which variables map to which plot
attributes. These mappings have reasonable default behaviors, but can be
modified through scales.
For example, you’ll occasionally want control over the location and specification of tick marks on the X or Y axis. These can be manipulated in scale_x_* and scale_y_* where * depends on the type of variable mapped to the x and y aesthetics (i.e. continuous vs discrete).
scale options:
weather_df %>%
ggplot(aes(x = tmin, y = tmax)) +
geom_point(aes(color = name), alpha = .5) +
labs(
title = "Temperature plot",
x = "Minimum daily temperature (C)",
y = "Maxiumum daily temperature (C)",
caption = "Data from the rnoaa package") +
scale_x_continuous(
breaks = c(-15, 0, 15),
labels = c("-15ºC", "0", "15"),
limits = c(-20, 30)) +
scale_y_continuous(
trans = "sqrt",
position = "right")## Warning in self$trans$transform(x): NaNs produced## Warning: Transformation introduced infinite values in continuous y-axis## Warning: Removed 90 rows containing missing values (geom_point).
different version - scale colors:
weather_df %>%
ggplot(aes(x = tmin, y = tmax)) +
geom_point(aes(color = name), alpha = .5) +
labs(
title = "Temperature plot",
x = "Minimum daily temperature (C)",
y = "Maxiumum daily temperature (C)",
caption = "Data from the rnoaa package") +
scale_color_hue(name = "Location", h = c(100, 300))## Warning: Removed 15 rows containing missing values (geom_point).
scale colors with viridis package:
ggp_temp_plot =
weather_df %>%
ggplot(aes(x = tmin, y = tmax)) +
geom_point(aes(color = name), alpha = .5) +
labs(
title = "Temperature plot",
x = "Minimum daily temperature (C)",
y = "Maxiumum daily temperature (C)",
caption = "Data from the rnoaa package"
) +
viridis::scale_color_viridis(
name = "Location",
discrete = TRUE
)
ggp_temp_plot## Warning: Removed 15 rows containing missing values (geom_point).
We used discrete = TRUE because the color aesthetic is mapped to a
discrete variable. In other cases (for example, when color mapped to
prcp) you can omit this argument to get a continuous color gradient. The
viridis::scale_fill_viridis() function is appropriate for the fill
aesthetic used in histograms, density plots, and elsewhere.
themes
Themes are used to modify non-data elements of a plot – they don’t change mappings or how data are render, but control things like background color and location of the the legend. Using themes can help with general plot appearance.
For example, I frequently change is the legend position. By default this is on the right of the graphic, but I like to shift it to the bottom to ensure the graphic takes up the available left-to-right space.
ggp_temp_plot +
theme(legend.position = "bottom")## Warning: Removed 15 rows containing missing values (geom_point).
Quick tip: legend.position = “none” will remove the legend. This is helpful when multiple plots use the same color scheme or when the legend is obnoxious for some other reason.
By default this is theme_gray; here’s theme_bw():
ggp_temp_plot +
theme_bw() +
theme(legend.position = "bottom")## Warning: Removed 15 rows containing missing values (geom_point).
theme_classic():
ggp_temp_plot +
theme_classic() +
theme(legend.position = "bottom")## Warning: Removed 15 rows containing missing values (geom_point).
here’s the Excel 2003 theme from ggthemes:
ggp_temp_plot +
ggthemes::theme_excel() +
theme(legend.position = "bottom")## Warning: Removed 15 rows containing missing values (geom_point).
The ordering of theme_bw() and theme() matters – theme() changes a particular element of the plot’s current “theme”. If you call theme to change the some element and then theme_bw(), the changes introduced by theme() are overwritten by theme_bw().
setting options
In addition to figure sizing, I include a few other figure preferences in global options declared at the outset of each .Rmd file (this code chunk just gets copy-and-pasted to the beginning of every new file).
library(tidyverse)
knitr::opts_chunk$set(
fig.width = 6,
fig.asp = .6,
out.width = "90%"
)
theme_set(theme_minimal() + theme(legend.position = "bottom"))
options(
ggplot2.continuous.colour = "viridis",
ggplot2.continuous.fill = "viridis"
)
scale_colour_discrete = scale_colour_viridis_d
scale_fill_discrete = scale_fill_viridis_ddata argument in geom_*
We’ve seen that where an aesthetic gets mapped to a variable matters – setting aes(color = name) in ggplot can yield different results than the same setting in geom_point(). This arises from the way that ggplot objects inherit aesthetic mappings, and it turns out there’s a similar thing with the data used to make a plot.
In a contrived example, we can split weather_df into separate datasets for Central Park and Waikiki. Then we use one in the ggplot() call and another in geom_line():
central_park =
weather_df %>%
filter(name == "CentralPark_NY")
waikiki =
weather_df %>%
filter(name == "Waikiki_HA")
ggplot(data = waikiki, aes(x = date, y = tmax, color = name)) +
geom_point() +
geom_line(data = central_park)## Warning: Removed 3 rows containing missing values (geom_point).
patchwork package
We’ve seen facetting as an approach to create the “same plot” for several levels of a categorical variable, and this can get you pretty far. Sometimes, though, you want to show two or three fundamentally different plots in the same graphic: you may want to juxtapose a scatterplot and a boxplot, or show scatterplots illustrating relationships between different variables. In this case, a solution is to create each of the panels you want separately and combine panels using tools in the patchwork package:
tmax_tmin_p =
weather_df %>%
ggplot(aes(x = tmax, y = tmin, color = name)) +
geom_point(alpha = .5) +
theme(legend.position = "none")
prcp_dens_p =
weather_df %>%
filter(prcp > 0) %>%
ggplot(aes(x = prcp, fill = name)) +
geom_density(alpha = .5) +
theme(legend.position = "none")
tmax_date_p =
weather_df %>%
ggplot(aes(x = date, y = tmax, color = name)) +
geom_point(alpha = .5) +
geom_smooth(se = FALSE) +
theme(legend.position = "bottom")
(tmax_tmin_p + prcp_dens_p) / tmax_date_p## Warning: Removed 15 rows containing missing values (geom_point).## `geom_smooth()` using method = 'loess' and formula 'y ~ x'## Warning: Removed 3 rows containing non-finite values (stat_smooth).## Warning: Removed 3 rows containing missing values (geom_point).
Data manipulation
reordering name by hand:
weather_df %>%
mutate(name = forcats::fct_relevel(name, c("Waikiki_HA", "CentralPark_NY", "Waterhole_WA"))) %>%
ggplot(aes(x = name, y = tmax)) +
geom_violin(aes(fill = name), color = "blue", alpha = .5) +
theme(legend.position = "bottom")## Warning: Removed 3 rows containing non-finite values (stat_ydensity).
A second example reorders name according to tmax values in each name:
weather_df %>%
mutate(name = forcats::fct_reorder(name, tmax)) %>%
ggplot(aes(x = name, y = tmax)) +
geom_violin(aes(fill = name), color = "blue", alpha = .5) +
theme(legend.position = "bottom")## Warning: Removed 3 rows containing non-finite values (stat_ydensity).
A more difficult situation relates to data tidyiness. Suppose I wanted to create a three-panel plot showing densities for tmax and tmin within each location. More concretely, I want to be able to facet panels across the name variable, and create separate densities for tmax and tmin in each panel. Unfortunately, weather_df isn’t organized in a way that makes this easy.
One solution would recognize that tmax and tmin are separate observation types of a shared temperature variable. With this understanding, it’s possible to tidy the weather_df and make the plot directly:
weather_df %>%
select(name, tmax, tmin) %>%
pivot_longer(
tmax:tmin,
names_to = "observation",
values_to = "temp") %>%
ggplot(aes(x = temp, fill = observation)) +
geom_density(alpha = .5) +
facet_grid(~name) +
viridis::scale_fill_viridis(discrete = TRUE)## Warning: Removed 18 rows containing non-finite values (stat_density).
pulse data example
pulse_data =
haven::read_sas("./data/public_pulse_data.sas7bdat") %>%
janitor::clean_names() %>%
pivot_longer(
bdi_score_bl:bdi_score_12m,
names_to = "visit",
names_prefix = "bdi_score_",
values_to = "bdi") %>%
select(id, visit, everything()) %>%
mutate(
visit = recode(visit, "bl" = "00m"),
visit = factor(visit, levels = str_c(c("00", "01", "06", "12"), "m"))) %>%
arrange(id, visit)
ggplot(pulse_data, aes(x = visit, y = bdi)) +
geom_boxplot()## Warning: Removed 879 rows containing non-finite values (stat_boxplot).
FAS data example
pup_data =
read_csv("./data/FAS_pups.csv", col_types = "ciiiii") %>%
janitor::clean_names() %>%
mutate(sex = recode(sex, `1` = "male", `2` = "female"))
litter_data =
read_csv("./data/FAS_litters.csv", col_types = "ccddiiii") %>%
janitor::clean_names() %>%
separate(group, into = c("dose", "day_of_tx"), sep = 3)
fas_data = left_join(pup_data, litter_data, by = "litter_number")
fas_data %>%
select(sex, dose, day_of_tx, pd_ears:pd_walk) %>%
pivot_longer(
pd_ears:pd_walk,
names_to = "outcome",
values_to = "pn_day") %>%
drop_na() %>%
mutate(outcome = forcats::fct_reorder(outcome, pn_day, median)) %>%
ggplot(aes(x = dose, y = pn_day)) +
geom_violin() +
facet_grid(day_of_tx ~ outcome)